Closed Techmind closed 8 years ago
Or maybe a mailing list or something.
Ok, probably a bug, added a spreadsheet to calculate difference (length between points) between ref & in, in & out, it became more, than original. Should always become less if i understand the purpose of ICP correctly(J47 > J23).
Spreadsheet: https://docs.google.com/spreadsheets/d/15IEtTkTp7KwdC0P09Vo7EiAYq7_vd25C9FoTAOUv5VU/edit#gid=1228734682
For now, we answer directly here as everything is at the same place.
What is your configuration file (something.yaml)? If you use the the default configuration (no yaml file), it is most likely that outliers rejection is removing some of your matches. The default configuration was intended to work for the example files, but it is suggested to always setup a yaml file for your own application.
The spreadsheet was created from default config(if y provide none), i think after playing with "knn" and epsilon i got a bit better results.
Probably default config should be VERY restrictive.
With this config:
readingDataPointsFilters:
- IdentityDataPointsFilter
referenceDataPointsFilters:
- IdentityDataPointsFilter
matcher:
KDTreeMatcher:
knn: 1
epsilon: 0.0000001
searchType: 0
outlierFilters:
- NullOutlierFilter
errorMinimizer:
PointToPointErrorMinimizer
transformationCheckers:
- CounterTransformationChecker:
maxIterationCount: 4
inspector:
NullInspector
logger:
NullLogger
i get this: http://s10.postimg.org/m799tc9ax/image.png, which is actually better.
Don't hesitate to use the VTKInspector to debug your solution. See this tutorial: https://github.com/ethz-asl/libpointmatcher/blob/master/doc/ICPIntro.md#a-real-icp-configuration
inspector:
VTKFileInspector:
dumpReading : 1
dumpReference : 1
dumpDataLinks : 1
The default configuration will always be wrong for an application or an other. That's why we tried to simplify the tuning of a solution.
Having weird results and unshure if bug or i am doing something wrong.
This should be 99% translation, created a gist instead: https://gist.github.com/Techmind/d52fd6d01756ced2fd68dbc5ab4da354